The authors of this survey propose a new way to categorize and understand AI agents that interact with users through command-line interfaces. They call these agents 'terminal agents' and argue that existing research has been scattered across various fields, including software engineering and computer-use research. The authors develop a framework for studying terminal-mediated agency, which includes seven dimensions of competence. They also highlight the importance of considering system architecture, interface design, and environment interaction when evaluating AI agents in command-line environments. --- Why it matters: This matters to researchers in AI because it provides a unified framework for understanding and comparing terminal-mediated agency across different domains. It can help identify best practices for designing and evaluating AI systems that interact with users through command-line interfaces. Source: https://arxiv.org/abs/2608.20485